/** Test that DataStorage and BlockPoolSliceStorage remove the failed volume after failure. */
@Test(timeout = 150000)
public void testFailedVolumeBeingRemovedFromDataNode()
throws InterruptedException, IOException, TimeoutException {
// The test uses DataNodeTestUtils#injectDataDirFailure() to simulate
// volume failures which is currently not supported on Windows.
assumeTrue(!Path.WINDOWS);
Path file1 = new Path("/test1");
DFSTestUtil.createFile(fs, file1, 1024, (short) 2, 1L);
DFSTestUtil.waitReplication(fs, file1, (short) 2);
File dn0Vol1 = new File(dataDir, "data" + (2 * 0 + 1));
DataNodeTestUtils.injectDataDirFailure(dn0Vol1);
DataNode dn0 = cluster.getDataNodes().get(0);
checkDiskErrorSync(dn0);
// Verify dn0Vol1 has been completely removed from DN0.
// 1. dn0Vol1 is removed from DataStorage.
DataStorage storage = dn0.getStorage();
assertEquals(1, storage.getNumStorageDirs());
for (int i = 0; i < storage.getNumStorageDirs(); i++) {
Storage.StorageDirectory sd = storage.getStorageDir(i);
assertFalse(sd.getRoot().getAbsolutePath().startsWith(dn0Vol1.getAbsolutePath()));
}
final String bpid = cluster.getNamesystem().getBlockPoolId();
BlockPoolSliceStorage bpsStorage = storage.getBPStorage(bpid);
assertEquals(1, bpsStorage.getNumStorageDirs());
for (int i = 0; i < bpsStorage.getNumStorageDirs(); i++) {
Storage.StorageDirectory sd = bpsStorage.getStorageDir(i);
assertFalse(sd.getRoot().getAbsolutePath().startsWith(dn0Vol1.getAbsolutePath()));
}
// 2. dn0Vol1 is removed from FsDataset
FsDatasetSpi<? extends FsVolumeSpi> data = dn0.getFSDataset();
try (FsDatasetSpi.FsVolumeReferences vols = data.getFsVolumeReferences()) {
for (FsVolumeSpi volume : vols) {
assertNotEquals(
new File(volume.getBasePath()).getAbsoluteFile(), dn0Vol1.getAbsoluteFile());
}
}
// 3. all blocks on dn0Vol1 have been removed.
for (ReplicaInfo replica : FsDatasetTestUtil.getReplicas(data, bpid)) {
assertNotNull(replica.getVolume());
assertNotEquals(
new File(replica.getVolume().getBasePath()).getAbsoluteFile(), dn0Vol1.getAbsoluteFile());
}
// 4. dn0Vol1 is not in DN0's configuration and dataDirs anymore.
String[] dataDirStrs = dn0.getConf().get(DFSConfigKeys.DFS_DATANODE_DATA_DIR_KEY).split(",");
assertEquals(1, dataDirStrs.length);
assertFalse(dataDirStrs[0].contains(dn0Vol1.getAbsolutePath()));
}
/** Test for {@link FsDatasetImpl#updateReplicaUnderRecovery(ExtendedBlock, long, long)} */
@Test
public void testUpdateReplicaUnderRecovery() throws IOException {
MiniDFSCluster cluster = null;
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
cluster.waitActive();
String bpid = cluster.getNamesystem().getBlockPoolId();
// create a file
DistributedFileSystem dfs = (DistributedFileSystem) cluster.getFileSystem();
String filestr = "/foo";
Path filepath = new Path(filestr);
DFSTestUtil.createFile(dfs, filepath, 1024L, (short) 3, 0L);
// get block info
final LocatedBlock locatedblock =
getLastLocatedBlock(DFSClientAdapter.getDFSClient(dfs).getNamenode(), filestr);
final DatanodeInfo[] datanodeinfo = locatedblock.getLocations();
Assert.assertTrue(datanodeinfo.length > 0);
// get DataNode and FSDataset objects
final DataNode datanode = cluster.getDataNode(datanodeinfo[0].getIpcPort());
Assert.assertTrue(datanode != null);
// initReplicaRecovery
final ExtendedBlock b = locatedblock.getBlock();
final long recoveryid = b.getGenerationStamp() + 1;
final long newlength = b.getNumBytes() - 1;
final FsDatasetSpi<?> fsdataset = DataNodeTestUtils.getFSDataset(datanode);
final ReplicaRecoveryInfo rri =
fsdataset.initReplicaRecovery(new RecoveringBlock(b, null, recoveryid));
// check replica
final ReplicaInfo replica =
FsDatasetTestUtil.fetchReplicaInfo(fsdataset, bpid, b.getBlockId());
Assert.assertEquals(ReplicaState.RUR, replica.getState());
// check meta data before update
FsDatasetImpl.checkReplicaFiles(replica);
// case "THIS IS NOT SUPPOSED TO HAPPEN"
// with (block length) != (stored replica's on disk length).
{
// create a block with same id and gs but different length.
final ExtendedBlock tmp =
new ExtendedBlock(
b.getBlockPoolId(),
rri.getBlockId(),
rri.getNumBytes() - 1,
rri.getGenerationStamp());
try {
// update should fail
fsdataset.updateReplicaUnderRecovery(tmp, recoveryid, newlength);
Assert.fail();
} catch (IOException ioe) {
System.out.println("GOOD: getting " + ioe);
}
}
// update
final String storageID =
fsdataset.updateReplicaUnderRecovery(
new ExtendedBlock(b.getBlockPoolId(), rri), recoveryid, newlength);
assertTrue(storageID != null);
} finally {
if (cluster != null) cluster.shutdown();
}
}